Accurate prediction of software defect density is crucial for improving software quality and reducing maintenance costs. While previous studies have applied deep learning models to this problem, there remains potential to enhance prediction accuracy through novel approaches. This paper introduces an advanced hybrid deep learning framework that integrates Convolutional Neural Networks (CNNs) with Recurrent Neural Networks (RNNs) and incorporates sophisticated feature engineering techniques. By leveraging a combination of temporal and spatial feature representations, the proposed method addresses limitations observed in traditional models. Experiments conducted on diverse software datasets demonstrate that the hybrid approach significantly improves prediction accuracy and robustness compared to conventional deep learning methods. This study offers valuable insights into optimizing defect prediction models and provides a foundation for future research in the domain.